Feedback Capture: Corrections Become Evals and Memory

When a person changes what an agent proposed, the change records what the agent wanted to do, what the person did, and the difference. Saving that difference with enough detail to use it again, and sending it where it can be acted on, is called feedback capture.

Without it, an employee can make the same correction every day for a year, and the agent never improves, because the model does not remember earlier calls. How does a correction made today prevent the same error next month?